"The product of mathematics is clarity and understanding. Not theorems, by themselves. Is there, for example any real reason that even such famous results as Fermat's Last Theorem, or the Poincaré conjecture, really matter? Their real importance is not in their specific statements, but their role in challenging our understanding, presenting challenges that led to mathematical developments that increased our understanding."
That's the product of math from the point of view of mathematicians. But is it the point of view of those funding math?
I suggest if one looks at the history of funding for mathematics and science, the product of these efforts is not understanding, but rather power. Funding went way up after WW2 when the war demonstrated that power flows from them. Math not only contributed to the scientific weapons of the way, but was directly used in operation planning (the birth of the field of Operations Research) as well as in cryptography.
The reason this matters is that AI is also a quintessential power-oriented technology. From the point of those providing the monetary lifeblood on which modern mathematical practice depends, the current math-AI discussion presents no issue worthy of concern.
There is more to math, than input (money) and output (power). Sure, there is some relation between applied sciences and how knowlegde can assist effecting world events.
But for the most part, math discovery relied more on human curiosity than on resources to "do math". Conversely, if people allocate lots of money to developing AI, that doesn't mean mathematicians have an obligation to take the money provide ROI to investors.
I mean, in real life it's a combination of both. Some money is for math as an exploration of our world that will never pay off. Some money is learning things that may pay off long after we're dead (planting trees so our great grandchildren have shade). Some money is for solving problems right now.
Getting funding can be quite difficult at times, so you'll see some portion of researchers (or mathematicians in this case) take the dollars they can get.
> That's the product of math from the point of view of mathematicians. But is it the point of view of those funding math?
Yes, and your examples are exactly examples of what the GP quote is talking about.
Of course people paying money want applications, which includes "power" in your kind of reductive framing (applications to war being only one of many types of applications, or we could redefine any gradient provided by expanded understanding as "power", in which case the choice of word just seems melodramatic).
What we've also learned over the centuries, a lot more clearly in the last few, is that seemingly pointless or applicationless understanding can very quickly become useful. This is why it's clearly worth still funding pure math.
Accelerationists may argue that the eroding of proper attribution and proof verification by humans is a meaningless short term struggle of a dying field.
Mathematics seems to be entering an era where human + machine maximizes performance, much like chess in the 1990s. However, imagine a future where even talented mathematicians are nothing but noise in the machine (as is the case in chess now). A future where AI generates and verifies proofs without humans in the loop. Where the mathematics may be beyond human comprehension.
In that future, does it matter that early career mathematicians are inhibited by these developments? Perhaps not. Programming faces the same issue. As AI crawls up the competence ladder, does it matter that fewer people have opportunities to develop the skillset of a senior engineer? Perhaps not.
Chess has set rules and is a closed world with a set objective. In maths, you make up the rules. For every clearly defined problem that everybody cares about and is not yet proven, there's someone who first who recognised the importance of that given problem and managed to define it clearly enough for it to be recognised as such.
There's also the separate, less glamorous issue that people don't want to talk about, which is proof reliability. [0] If you have systems to help you formalise the problems and leave an algorithm or AI or whatever solve it in a verifiable way, that's a win for both the mathematicians and the rest of the world.
The deeper question is whether AI can replace the human role in deciding what mathematics should be done and what concepts matter. If that's automated, then yeah, we're screwed.
Chess is a game where we play and observe for the sheer joy of mastery. Mathematics has some of that “done for its own sake” aspect but not as cleanly. There’s always the chance that some applied mathematics might help something in science or engineering. Whether we care more about process vs outcome is now more important than ever before.
(This doesn't disprove your point because it seems like it should be trivial to make but interestingly)
I don't think there's any computer system which autonomously come up with new directions in openings. As far as I know; a GM looks at stockfish's evaluation of the top x moves and analyzes one that hasn't been played a lot etc.
> imagine a future where even talented mathematicians are nothing but noise
That would be AGI. My conjecture is that LLMs alone are not enough for that future. They are incredible, but AGI needs other breakthroughs.
In that sense, I think math is very different from chess or Go. Chess and Go are complete-information games with fixed rules and a fixed board. Math is open-ended.
Most of the arguments here feel like gate keeping and resistance to change. I didn't see any arguments that were directly about advancing the state of knowledge of math.
“Current automated techniques can produce plausible but unreliable (or even incorrect) arguments which are difficult to distinguish from correct mathematical proofs.”
That seems like a problem for mathematics with or without AI.
Isn’t this a problem with human proofs as well?
“Many current models are also built on data obtained by systematically exploiting licenses and access arrangements that were not made with artificial intelligence in mind, or indeed by simply violating copyright protections”
Copyright? The copyright arguments have been hard to make in domains where copyright is much stronger, mathematical knowledge isn’t even subject to copyright.
“Technologies which affect the way in which mathematics is practiced may disturb the current system of incentives”
Resistance to change again.
“Proper evaluation is endangered if results are communicated through informal channels”
There is some of that but I wouldn't call it gatekeeping. Universities lately promote citations and publications so there's a sense that results are all that matters. Results matter, yes, but there's a human side too where we're kind of asking about human creativity and ability. To me an appropriate analogy is in climbing Mt. Everest. Proving something, or even writing a thesis, is like climbing Mt. Everest. A lot of the value is actually in the effort you put into it. You could take a helicopter ride up to the top and then climb a few steps and claim "You climbed to the peak of Everest". That's like using AI. But if you asked them about what it was like, how they prepared, etc. their answer would not be helpful. So I think there is a lot of value in the journey itself and outsourcing all this to AI would destroy the human part of it.
It's completely antithetical to the whole enterprise to hide anything these researchers produce behind a paywall. Id be glad to see that go.
> or even writing a thesis, is like climbing Mt. Everest. A lot of the value is actually in the effort you put into it.
As an analogy, in the music industry, if you need a jingle written, you wouldn't care if someone spent five minutes or five years writing it. AI is now filling that formulaic space very well. It won't replace the top end of humans output but it completely outdoes all the boilerplate stuff humans take an age creating
Human proofs are themselves a kind of a proof of work. They certainly write flawed proofs, but you can expect a human author of a paper to have put in more effort--substantially more--than the human reader needs to verify it. Arguably, this asymmetry disappears for generated proofs.
Automated theorem provers help a bit here, but they don't eliminate the human verification cost.
Can't all proofs be eventually broken down into their fundamental pieces and then it's clear as day if it's right or wrong? Seems like a proof would be the best place to determine if an AI is right or not because the output is either right or wrong, there's no subjectivity and the, now common, excuse "well a human would have done the same" won't hold up.
> Can't all proofs be eventually broken down into their fundamental pieces and then it's clear as day if it's right or wrong?
You’d think so, but not really. There are mathematical structures which are unimaginably huge but have little if any reducible structure. For example, in algebra, one of the most basic structures is a Group. When trying to understand a group, one of the most important tools is to break a group into chunks using what’s called a “normal subgroup”. However it turns out that there are some absolutely enormous groups that are “simple” (ie have no normal subgroups). So, there is a set of 26 of these known as the “sporadic simple groups” that just don’t fit any kind of pattern. Proving results about these has proved very difficult because they can’t be broken down (they have no normal subgroups) and by definition just don’t fit any kind of other pattern. One of these, the “monster” group has approximately 8x10^53 members. So you have a set that is unimaginably massive and has very little internal structure as it is “simple” and so can’t be broken down further.
The proof that there are 26 of these sporadic simple groups is part of the theorem known as the classification of finite simple groups, sometimes known as the “Enormous Theorem”.[1] It took over 100 mathematicians nearly 50 years and resulted in hundreds of papers. Even with that many mathematicians involved, there were still errors and revisions needed to the original proof. Some of the original authors are gradually publishing a somewhat simplified version of the proof but it’s still a massive effort.
Generally, yes, but once broken down you end up with a large number of items that individually each is obviously true, so you know the combined statement Is true, but you don't find out if it is saying what you think it is saying.
In combining the parts you have the correct answer to a question, but is it that question you want to know?
Consider a proof that in the future all people will be happy.
You can methodically show this to be true but at the same time inadvertently include a proof that the number of people in the future will be zero.
It doesn't make the claim wrong, it stays undoubtedly true. It's just not what you assume it means.
It is, but it is somewhat worse for machine-generated proofs, especially when the proof is very long and was done by brute force (eg the 4 colour map theorem[1] is the famous example), or depends on a lot of niche results in disparate areas (which LLMs are wont to sometimes do).
Even when the proof is produced by the llm in a formal system like Lean4 it may not be “honest”[2] and it can be hard to tell if the proof is very long and complex and especially if it includes highly specialized results from lots of different areas of maths. Llms can (and do) do this just fine, but for a human proof that would require a team each of which was specialized in a particular area. Those people are more likely to be able to cross-check each other.
[2] An “honest” proof may contain bugs or errors but it does not constitute a deliberate attack on the proof system or the math libraries it uses. Systems like Lean aim to not incorrectly validate an honest proof with mistakes but don’t guarantee anything in the case of a proof being dishonest. This is the sense used here https://lean-lang.org/doc/reference/latest/ValidatingProofs/ .
Your list is cherry picked from the list of "potential threats" to the values of the mathematical research community identified by this document. They aren't criticisms or absolute statements, they're literally a list of potential new problems for the future of mathematical research, and they all seem reasonable to me, if not all at the same levels of magnitude or plausibility.
Notably you don't seem to be looking at either the list of identified values or their recommendations to researchers in their use of LLMs, which would seem much more important to engage with in any non-shallow dismissal of the document as "feel[ing] like gate keeping and resistance to change".
It's also kind of a bad look (and actively harmful for discourse) for people working on AI to be so dismissive of fields actively engaging with how their field is changing due to AI. I haven't seen any other field engaging this actively with its possible futures, have you? Usually we seem to only get some extreme of over-hyped utopia, doomerism, or dismissal of everything as slop.
Interesting document. Especially interesting compared to the Bourbaki movement from a century ago, which was much more focused on universality and correctness, and much less focused on process and attribution (in fact, demanded anonymity).
I consider attribution very important, less because it were associated with some merits for the humans quoted as the sources of certain new ideas, but more because it facilitates going backwards to the historical documents where those ideas were first introduced.
At least for me, in many cases I have achieved a much better understanding of various things after I studied the historical development of the ideas related to them.
Therefore I agree with the point "2." at "Potential Threats". For me a novel mathematical demonstration that is not presented in a way which disentangles its really new elements from the previously known elements, by proper quotation of all relevant older sources, has a value that is many times lower than that of a demonstration with proper attributions.
> Terence Tao - Professor, University of California, Los Angeles
> This has been the result of months of community input about the fundamental values and goals of the mathematical community. In retrospect, these were questions we should have been systematically discussing years ago, but in any event the exercise was extremely valuable, and the end result is excellent. I wholeheartedly endorse the statements and recommendations in this declaration.
>I support this declaration. I have one small comment: the document notes that "Technologies which affect the way in which mathematics is practiced may disturb the current system of incentives." The current system of incentives seriously is flawed in many ways, and I don't think maintaining the status quo should be our goal. However, we should work to improve it, not let it be corrupted by outside forces, as has already been done for decades by university administrators, journal oligopolies, etc.
Let's assume that there is an advantage to using llms. E.g. the use of an llm lends a competitive advantage in a given field.
There is no moral or ethical obligation to disclose tool use. The disclosure in of itself presents an asymmetric disadvantage to the disclosee. Especially in this charged environment where large swathes of people are champing at the bit to discredit or diminish any effort that leverages these tools.
This system incentivizes people to hide tool use to gain a competitive advantage.
This moralistic grandstanding will be seen as a reactionary movement of people trying to cope with transformative technology.
Lie about tool use, don't admit it. Use it as you see fit and rely on your taste, expertise and best judgement.
> Technologies that draw extensively on the published mathematical commons undermine the traditional system of attribution.
This just feels like something that has always been true. Defending attribution in this way feels more like a panicked gatekeeping rather than something valuable and principled. I’m a bit disappointed to see people like Terence Tao endorse this.
A career in mathematics is very unlikely to deliver fortune, fame, or even respect. Attribution is in many cases the only way we know that someone else thought our work was valuable.
The bar for being cited has never been lower in the history of mathematics: getting an arXiv account is easier than getting a lobste.rs account (at least by my experience, for I have the former but not the latter).
Tao knows what he is talking about, he was an early adopter of LLMs for proof-generation.
yes, that's where a conference was held that kickstarted the group that drafted this declaration.
> In September 2025 the Lorentz Center at Leiden University in the Netherlands hosted a conference entitled Mechanization and Mathematical Research. The around 60 participants from 10 countries comprised mathematicians, computer scientists, philosophers, historians and social scientists, including those with experience in industry and in government.
Who exactly is the implied underdog "you" that we so desperately want to win here-- do you mean those poor struggling $xx billion companies or current US government apparently beholden to them?
The potential threats section reads like panic, rather than a critique of AI. I can see where #2 has some legs, if I thought tradition was sacrosanct.
1. AI proofs might be incorrect and difficult to demonstrate why. This implies they are not like human proofs in these qualities.
2. AI proofs are difficult to attribute correctly, because they don't follow established traditions. Nothing to do with the math, but ok.
3. Mathematicians without AI (for political or practical reasons) will not necessarily be able to participate in AI-assisted research. This history of Mathematics is littered with people having uneven access.
4. People/orgs are publishing that AI found things are fact before they are properly evaluated. Same issue.
5. All these things are bad, because AI might muddy the field with lots of unknowns.
This appears to be a very bad faith post that intentionally misrepresents what is being said.
1. pertains to the quantity of output adding stress to review processes; LLMs can feasibly produce a million plausible but incorrect 'proofs' in the time that a human can produce one. We already see this effect in software development, with bug bounty programs shutting down and open-source software rejecting AI contributions or closing altogether because LLMs flood review channels with an amount of spam for which there is no sufficient amount of human bandwidth to handle.
2. is nothing about "following established traditions" but rather the general concept of crediting people for their prior work, unless you think that "not plagiarising" is a trifling established tradition.
3. is more or less accurate to the point they made, but "it has historically been this way" isn't a compelling justification for "it should always be this way and also it's okay if it gets worse"
4. An existing issue being made 100x more common is a point worth bringing attention to even if it already existed, actually
5. said nothing that could possibly be interpreted in the vein of "muddying the field with lots of unknowns" at all. Point 5 was actually about economic incentives and the risk of mathematic research becoming beholden to tech monopolies
>2. is nothing about "following established traditions" but rather the general concept of crediting people for their prior work, unless you think that "not plagiarising" is a trifling established tradition.
But that is the nature of establishment, when something is a sufficiently firmly established tradition, people see it as a truism.
Crediting people is a social convention. Plagiarism is a social construct. It can be useful, in many areas of science, to reference to support your arguments. This is less important in proofs, because a proof is a proof, but references aid in understanding.
These are all reasons to reference and attribute that benefit the writer, and could be done voluntarily. The notion of a duty to reference or attribute has no impact on the validity of the claims being made. It is a collective decision to proportion prestige.
Turning the duty to do so into an unquestioned truism means it has to be done regardless of whether it accurately represents any property of merit.
There are many instances where prestige delivered grossly mismatches what an impartial observer would consider a fair balance of effort and ability.
We should at least recognise that this is so because we have chosen to let it be this way.
In the research papers of any branch of science and technology, crediting the previous work is not some social convention, but it is very useful information for readers.
Despite the fact that there are some dishonest researchers, who attempt to create the illusion that their work is revolutionary and does not owe anything or only very little to predecessors, the truth is that almost anything novel that is published today contains only a few percent of truly new ideas that are grafted on a big body that combines many results extracted for older works.
If a new research paper separates properly its new elements from the old elements, which are properly attributed, enabling the search of the original sources, that can help a lot the readers to understand it.
I consider the fact that when you ask a question to an LLM, it is unable to accurately provide the source of the answer, as their greatest defect.
The actual thesis of point 2 is about plagiarism, and the thesis would remain the same if the sentence you quoted were removed completely. Your portrayal of it moves the out-of-context snippet to the forefont of the argument and makes it sound like an issue of "tradition for tradition's sake" or something similarly indefensible, but you refuse to engage with the real argument being made, hence why I suspect you are acting in bad faith. Are you suggesting that not attributing credit to work you've copied from is the way things should be going forward? If you are, then argue that point and make it earnestly. Instead you continue to avoid any substantial discussion of the points raised and only went for a cheap "gotcha".
I do not agree that the thesis of point 2 is about plagiarism.
Plagiarism is a minor concern in research papers. Plagiarism means that the paper has 0% new content and everything in it is extracted from older works and combined.
However, even very valuable research papers may have only 1 new idea or only a few new ideas, which can be applied only together with a much greater amount of old knowledge, which might be mentioned explicitly in the paper, with proper attribution, enabling the reader to search the original sources, or the necessary old knowledge may be implicit, the writers assuming it to be known by the readers.
It is important that the writers should not imply that some old idea is new, by omitting to attribute it, but much more important is that when they mention some ideas as being old, they must also provide sufficient information about the sources of those old ideas, so that the readers will be able to access them.
It's worth remembering Thurston's essay on mathoverflow (https://mathoverflow.net/questions/43690/whats-a-mathematici...):
"The product of mathematics is clarity and understanding. Not theorems, by themselves. Is there, for example any real reason that even such famous results as Fermat's Last Theorem, or the Poincaré conjecture, really matter? Their real importance is not in their specific statements, but their role in challenging our understanding, presenting challenges that led to mathematical developments that increased our understanding."
That's the product of math from the point of view of mathematicians. But is it the point of view of those funding math?
I suggest if one looks at the history of funding for mathematics and science, the product of these efforts is not understanding, but rather power. Funding went way up after WW2 when the war demonstrated that power flows from them. Math not only contributed to the scientific weapons of the way, but was directly used in operation planning (the birth of the field of Operations Research) as well as in cryptography.
The reason this matters is that AI is also a quintessential power-oriented technology. From the point of those providing the monetary lifeblood on which modern mathematical practice depends, the current math-AI discussion presents no issue worthy of concern.
Power depends on understanding - Seeing a larger scale view of what is happening as opposed to an arbitrary sequence of manipulations.
The foundations of the WW2 technologies you cite were dependent on previous theoretical efforts (ex:relativity) to develop a good understanding.
Without understanding, you get brittle demos which fail as the environment or problem description changes.
There is more to math, than input (money) and output (power). Sure, there is some relation between applied sciences and how knowlegde can assist effecting world events.
But for the most part, math discovery relied more on human curiosity than on resources to "do math". Conversely, if people allocate lots of money to developing AI, that doesn't mean mathematicians have an obligation to take the money provide ROI to investors.
I mean, in real life it's a combination of both. Some money is for math as an exploration of our world that will never pay off. Some money is learning things that may pay off long after we're dead (planting trees so our great grandchildren have shade). Some money is for solving problems right now.
Getting funding can be quite difficult at times, so you'll see some portion of researchers (or mathematicians in this case) take the dollars they can get.
> That's the product of math from the point of view of mathematicians. But is it the point of view of those funding math?
Yes, and your examples are exactly examples of what the GP quote is talking about.
Of course people paying money want applications, which includes "power" in your kind of reductive framing (applications to war being only one of many types of applications, or we could redefine any gradient provided by expanded understanding as "power", in which case the choice of word just seems melodramatic).
What we've also learned over the centuries, a lot more clearly in the last few, is that seemingly pointless or applicationless understanding can very quickly become useful. This is why it's clearly worth still funding pure math.
Accelerationists may argue that the eroding of proper attribution and proof verification by humans is a meaningless short term struggle of a dying field.
Mathematics seems to be entering an era where human + machine maximizes performance, much like chess in the 1990s. However, imagine a future where even talented mathematicians are nothing but noise in the machine (as is the case in chess now). A future where AI generates and verifies proofs without humans in the loop. Where the mathematics may be beyond human comprehension.
In that future, does it matter that early career mathematicians are inhibited by these developments? Perhaps not. Programming faces the same issue. As AI crawls up the competence ladder, does it matter that fewer people have opportunities to develop the skillset of a senior engineer? Perhaps not.
Chess has set rules and is a closed world with a set objective. In maths, you make up the rules. For every clearly defined problem that everybody cares about and is not yet proven, there's someone who first who recognised the importance of that given problem and managed to define it clearly enough for it to be recognised as such.
There's also the separate, less glamorous issue that people don't want to talk about, which is proof reliability. [0] If you have systems to help you formalise the problems and leave an algorithm or AI or whatever solve it in a verifiable way, that's a win for both the mathematicians and the rest of the world.
The deeper question is whether AI can replace the human role in deciding what mathematics should be done and what concepts matter. If that's automated, then yeah, we're screwed.
[0] https://lamport.azurewebsites.net/tla/proof-statistics.html
> However, imagine a future where even talented mathematicians are nothing but noise in the machine (as is the case in chess now).
Isn't chess more popular than ever? Ai dominating the game didnt seen to matter
Chess is a game where we play and observe for the sheer joy of mastery. Mathematics has some of that “done for its own sake” aspect but not as cleanly. There’s always the chance that some applied mathematics might help something in science or engineering. Whether we care more about process vs outcome is now more important than ever before.
Sure, but if we continue that analogy it does mean that there will be no human contributions to frontier mathematics.
(This doesn't disprove your point because it seems like it should be trivial to make but interestingly)
I don't think there's any computer system which autonomously come up with new directions in openings. As far as I know; a GM looks at stockfish's evaluation of the top x moves and analyzes one that hasn't been played a lot etc.
> imagine a future where even talented mathematicians are nothing but noise
That would be AGI. My conjecture is that LLMs alone are not enough for that future. They are incredible, but AGI needs other breakthroughs.
In that sense, I think math is very different from chess or Go. Chess and Go are complete-information games with fixed rules and a fixed board. Math is open-ended.
Most of the arguments here feel like gate keeping and resistance to change. I didn't see any arguments that were directly about advancing the state of knowledge of math.
“Current automated techniques can produce plausible but unreliable (or even incorrect) arguments which are difficult to distinguish from correct mathematical proofs.”
That seems like a problem for mathematics with or without AI.
Isn’t this a problem with human proofs as well?
“Many current models are also built on data obtained by systematically exploiting licenses and access arrangements that were not made with artificial intelligence in mind, or indeed by simply violating copyright protections”
Copyright? The copyright arguments have been hard to make in domains where copyright is much stronger, mathematical knowledge isn’t even subject to copyright.
“Technologies which affect the way in which mathematics is practiced may disturb the current system of incentives”
Resistance to change again.
“Proper evaluation is endangered if results are communicated through informal channels”
Gatekeeping again.
There is some of that but I wouldn't call it gatekeeping. Universities lately promote citations and publications so there's a sense that results are all that matters. Results matter, yes, but there's a human side too where we're kind of asking about human creativity and ability. To me an appropriate analogy is in climbing Mt. Everest. Proving something, or even writing a thesis, is like climbing Mt. Everest. A lot of the value is actually in the effort you put into it. You could take a helicopter ride up to the top and then climb a few steps and claim "You climbed to the peak of Everest". That's like using AI. But if you asked them about what it was like, how they prepared, etc. their answer would not be helpful. So I think there is a lot of value in the journey itself and outsourcing all this to AI would destroy the human part of it.
It's completely antithetical to the whole enterprise to hide anything these researchers produce behind a paywall. Id be glad to see that go.
> or even writing a thesis, is like climbing Mt. Everest. A lot of the value is actually in the effort you put into it.
As an analogy, in the music industry, if you need a jingle written, you wouldn't care if someone spent five minutes or five years writing it. AI is now filling that formulaic space very well. It won't replace the top end of humans output but it completely outdoes all the boilerplate stuff humans take an age creating
> Isn’t this a problem with human proofs as well?
Human proofs are themselves a kind of a proof of work. They certainly write flawed proofs, but you can expect a human author of a paper to have put in more effort--substantially more--than the human reader needs to verify it. Arguably, this asymmetry disappears for generated proofs.
Automated theorem provers help a bit here, but they don't eliminate the human verification cost.
You aren’t really engaging with the substance or heart of the post, and your reading feels a bit knee-jerky and bad-faith to me.
Can't all proofs be eventually broken down into their fundamental pieces and then it's clear as day if it's right or wrong? Seems like a proof would be the best place to determine if an AI is right or not because the output is either right or wrong, there's no subjectivity and the, now common, excuse "well a human would have done the same" won't hold up.
The proof that there are 26 of these sporadic simple groups is part of the theorem known as the classification of finite simple groups, sometimes known as the “Enormous Theorem”.[1] It took over 100 mathematicians nearly 50 years and resulted in hundreds of papers. Even with that many mathematicians involved, there were still errors and revisions needed to the original proof. Some of the original authors are gradually publishing a somewhat simplified version of the proof but it’s still a massive effort.
[1] https://en.wikipedia.org/wiki/Classification_of_finite_simpl...
Generally, yes, but once broken down you end up with a large number of items that individually each is obviously true, so you know the combined statement Is true, but you don't find out if it is saying what you think it is saying.
In combining the parts you have the correct answer to a question, but is it that question you want to know?
Consider a proof that in the future all people will be happy.
You can methodically show this to be true but at the same time inadvertently include a proof that the number of people in the future will be zero.
It doesn't make the claim wrong, it stays undoubtedly true. It's just not what you assume it means.
It is, but it is somewhat worse for machine-generated proofs, especially when the proof is very long and was done by brute force (eg the 4 colour map theorem[1] is the famous example), or depends on a lot of niche results in disparate areas (which LLMs are wont to sometimes do).
Even when the proof is produced by the llm in a formal system like Lean4 it may not be “honest”[2] and it can be hard to tell if the proof is very long and complex and especially if it includes highly specialized results from lots of different areas of maths. Llms can (and do) do this just fine, but for a human proof that would require a team each of which was specialized in a particular area. Those people are more likely to be able to cross-check each other.
[1] https://pubs.ams.org/ebooks/conm/098/ and https://en.wikipedia.org/wiki/Four_color_theorem
[2] An “honest” proof may contain bugs or errors but it does not constitute a deliberate attack on the proof system or the math libraries it uses. Systems like Lean aim to not incorrectly validate an honest proof with mistakes but don’t guarantee anything in the case of a proof being dishonest. This is the sense used here https://lean-lang.org/doc/reference/latest/ValidatingProofs/ .
https://lean-lang.org/doc/reference/latest/ValidatingProofs/...
Your list is cherry picked from the list of "potential threats" to the values of the mathematical research community identified by this document. They aren't criticisms or absolute statements, they're literally a list of potential new problems for the future of mathematical research, and they all seem reasonable to me, if not all at the same levels of magnitude or plausibility.
Notably you don't seem to be looking at either the list of identified values or their recommendations to researchers in their use of LLMs, which would seem much more important to engage with in any non-shallow dismissal of the document as "feel[ing] like gate keeping and resistance to change".
It's also kind of a bad look (and actively harmful for discourse) for people working on AI to be so dismissive of fields actively engaging with how their field is changing due to AI. I haven't seen any other field engaging this actively with its possible futures, have you? Usually we seem to only get some extreme of over-hyped utopia, doomerism, or dismissal of everything as slop.
Interesting document. Especially interesting compared to the Bourbaki movement from a century ago, which was much more focused on universality and correctness, and much less focused on process and attribution (in fact, demanded anonymity).
I consider attribution very important, less because it were associated with some merits for the humans quoted as the sources of certain new ideas, but more because it facilitates going backwards to the historical documents where those ideas were first introduced.
At least for me, in many cases I have achieved a much better understanding of various things after I studied the historical development of the ideas related to them.
Therefore I agree with the point "2." at "Potential Threats". For me a novel mathematical demonstration that is not presented in a way which disentangles its really new elements from the previously known elements, by proper quotation of all relevant older sources, has a value that is many times lower than that of a demonstration with proper attributions.
Ever since I found out the french chopped their kings head off, I've had a quiet respect for those crazy frogs
> Terence Tao - Professor, University of California, Los Angeles
> This has been the result of months of community input about the fundamental values and goals of the mathematical community. In retrospect, these were questions we should have been systematically discussing years ago, but in any event the exercise was extremely valuable, and the end result is excellent. I wholeheartedly endorse the statements and recommendations in this declaration.
John Carlos Baez:
>I support this declaration. I have one small comment: the document notes that "Technologies which affect the way in which mathematics is practiced may disturb the current system of incentives." The current system of incentives seriously is flawed in many ways, and I don't think maintaining the status quo should be our goal. However, we should work to improve it, not let it be corrupted by outside forces, as has already been done for decades by university administrators, journal oligopolies, etc.
> Recommendations for policymakers in government and elsewhere
Let's assume that there is an advantage to using llms. E.g. the use of an llm lends a competitive advantage in a given field.
There is no moral or ethical obligation to disclose tool use. The disclosure in of itself presents an asymmetric disadvantage to the disclosee. Especially in this charged environment where large swathes of people are champing at the bit to discredit or diminish any effort that leverages these tools.
This system incentivizes people to hide tool use to gain a competitive advantage.
This moralistic grandstanding will be seen as a reactionary movement of people trying to cope with transformative technology.
Lie about tool use, don't admit it. Use it as you see fit and rely on your taste, expertise and best judgement.
1. 'incentives' hmm
"Now, here, you see, it takes all the running you can do, to keep in the same place."
(Leiden being the town in the Netherlands where Leiden University is.)
Arguably the most beautiful town in the Netherlands
With a gorgeous botanical garden and great cafes along the canals
Beautiful area. I lived there for a time during the pandemic and I really enjoyed walking the canals.
So they are recommendations. At global level.
In a word, they're cooked
> Technologies that draw extensively on the published mathematical commons undermine the traditional system of attribution.
This just feels like something that has always been true. Defending attribution in this way feels more like a panicked gatekeeping rather than something valuable and principled. I’m a bit disappointed to see people like Terence Tao endorse this.
A career in mathematics is very unlikely to deliver fortune, fame, or even respect. Attribution is in many cases the only way we know that someone else thought our work was valuable.
The bar for being cited has never been lower in the history of mathematics: getting an arXiv account is easier than getting a lobste.rs account (at least by my experience, for I have the former but not the latter).
Tao knows what he is talking about, he was an early adopter of LLMs for proof-generation.
Is there a connection to Leiden, NL?
yes, that's where a conference was held that kickstarted the group that drafted this declaration.
> In September 2025 the Lorentz Center at Leiden University in the Netherlands hosted a conference entitled Mechanization and Mathematical Research. The around 60 participants from 10 countries comprised mathematicians, computer scientists, philosophers, historians and social scientists, including those with experience in industry and in government.
These type of questions are the reason why LLMs will take your job :-)
Questions are in the spirit of community. It isn't listed on the about page. If you rely on LLMs for everything you will never interact with people.
1. first they ignore you <<<< GPT-4 can barely add too numbers without making a mistake
2. then they laugh at you <<<< the International Math Olympiad is basically just high school math
3. then they fight you <<<< this declaration
4. then you win
Who exactly is the implied underdog "you" that we so desperately want to win here-- do you mean those poor struggling $xx billion companies or current US government apparently beholden to them?
Let's not sell them short, they're closer to $xxxx billion.
> add too numbers
Did you do this on purpose to anger both Mathematicians and keen spellers?
Probably. I do it 2.
4. then you write empty cliches
The potential threats section reads like panic, rather than a critique of AI. I can see where #2 has some legs, if I thought tradition was sacrosanct.
1. AI proofs might be incorrect and difficult to demonstrate why. This implies they are not like human proofs in these qualities.
2. AI proofs are difficult to attribute correctly, because they don't follow established traditions. Nothing to do with the math, but ok.
3. Mathematicians without AI (for political or practical reasons) will not necessarily be able to participate in AI-assisted research. This history of Mathematics is littered with people having uneven access.
4. People/orgs are publishing that AI found things are fact before they are properly evaluated. Same issue.
5. All these things are bad, because AI might muddy the field with lots of unknowns.
This appears to be a very bad faith post that intentionally misrepresents what is being said.
1. pertains to the quantity of output adding stress to review processes; LLMs can feasibly produce a million plausible but incorrect 'proofs' in the time that a human can produce one. We already see this effect in software development, with bug bounty programs shutting down and open-source software rejecting AI contributions or closing altogether because LLMs flood review channels with an amount of spam for which there is no sufficient amount of human bandwidth to handle.
2. is nothing about "following established traditions" but rather the general concept of crediting people for their prior work, unless you think that "not plagiarising" is a trifling established tradition.
3. is more or less accurate to the point they made, but "it has historically been this way" isn't a compelling justification for "it should always be this way and also it's okay if it gets worse"
4. An existing issue being made 100x more common is a point worth bringing attention to even if it already existed, actually
5. said nothing that could possibly be interpreted in the vein of "muddying the field with lots of unknowns" at all. Point 5 was actually about economic incentives and the risk of mathematic research becoming beholden to tech monopolies
>2. is nothing about "following established traditions" but rather the general concept of crediting people for their prior work, unless you think that "not plagiarising" is a trifling established tradition.
But that is the nature of establishment, when something is a sufficiently firmly established tradition, people see it as a truism.
Crediting people is a social convention. Plagiarism is a social construct. It can be useful, in many areas of science, to reference to support your arguments. This is less important in proofs, because a proof is a proof, but references aid in understanding.
These are all reasons to reference and attribute that benefit the writer, and could be done voluntarily. The notion of a duty to reference or attribute has no impact on the validity of the claims being made. It is a collective decision to proportion prestige.
Turning the duty to do so into an unquestioned truism means it has to be done regardless of whether it accurately represents any property of merit.
There are many instances where prestige delivered grossly mismatches what an impartial observer would consider a fair balance of effort and ability.
We should at least recognise that this is so because we have chosen to let it be this way.
In the research papers of any branch of science and technology, crediting the previous work is not some social convention, but it is very useful information for readers.
Despite the fact that there are some dishonest researchers, who attempt to create the illusion that their work is revolutionary and does not owe anything or only very little to predecessors, the truth is that almost anything novel that is published today contains only a few percent of truly new ideas that are grafted on a big body that combines many results extracted for older works.
If a new research paper separates properly its new elements from the old elements, which are properly attributed, enabling the search of the original sources, that can help a lot the readers to understand it.
I consider the fact that when you ask a question to an LLM, it is unable to accurately provide the source of the answer, as their greatest defect.
I'm not sure it's constructive to explain our differences, point by point. eg
> 2. is nothing about "following established traditions"
> undermine the traditional system of attribution
Literally does.
Suffice to say, I find your interpretations to be surprising and disconnected and it has not changed my views.
The actual thesis of point 2 is about plagiarism, and the thesis would remain the same if the sentence you quoted were removed completely. Your portrayal of it moves the out-of-context snippet to the forefont of the argument and makes it sound like an issue of "tradition for tradition's sake" or something similarly indefensible, but you refuse to engage with the real argument being made, hence why I suspect you are acting in bad faith. Are you suggesting that not attributing credit to work you've copied from is the way things should be going forward? If you are, then argue that point and make it earnestly. Instead you continue to avoid any substantial discussion of the points raised and only went for a cheap "gotcha".
I do not agree that the thesis of point 2 is about plagiarism.
Plagiarism is a minor concern in research papers. Plagiarism means that the paper has 0% new content and everything in it is extracted from older works and combined.
However, even very valuable research papers may have only 1 new idea or only a few new ideas, which can be applied only together with a much greater amount of old knowledge, which might be mentioned explicitly in the paper, with proper attribution, enabling the reader to search the original sources, or the necessary old knowledge may be implicit, the writers assuming it to be known by the readers.
It is important that the writers should not imply that some old idea is new, by omitting to attribute it, but much more important is that when they mention some ideas as being old, they must also provide sufficient information about the sources of those old ideas, so that the readers will be able to access them.